Probabilistic optimized ranking for multimedia semantic concept detection via RVM

摘要: We present a probabilistic ranking-driven classifier for the detection of video semantic concept, such as airplane, building, etc. Most existing concept systems utilize Support Vector Machines (SVM) to perform and ranking retrieved shots. However, margin maximization principle SVM does not optimization but merely classification error minimization. To tackle this problem, we exploit sparse Bayesian kernel model, namely relevance vector machine (RVM), detection. Based on automatic determination principle, RVM outputs posterior prediction concepts. This inference output is optimal target shots, according Probabilistic Ranking Principle. The probability individual uni-modal features also facilitates fusion multi-modal evidences minimize Bayes risk. demonstrate both theoretically empirically that outperforms testings TRECVID 07 dataset show produces statically significant improvements in MAP scores over SVM-based methods.